Algorithmic Graph Theory
Part III
Perfect Graphs and Their Subclasses
Martin [email protected]
University of Primorska, Koper, Slovenia
Dipartimento di InformaticaUniversita degli Studi di Verona, March 2013
1 / 55
What we’ll do
1 THE BASICS.2 PERFECT GRAPHS.3 COGRAPHS.4 CHORDAL GRAPHS.5 SPLIT GRAPHS.6 THRESHOLD GRAPHS.7 INTERVAL GRAPHS.
2 / 55
THE BASICS.
2 / 55
Induced Subgraphs
Recall:
Definition
Given two graphs G = (V ,E) and G′ = (V ′,E ′),we say that G is an induced subgraph of G′ ifV ⊆ V ′
and E = {uv ∈ E ′ : u, v ∈ V}.
Equivalently: G can be obtained from G′ by deleting vertices.
Notation: G < G′
3 / 55
Hereditary Graph Properties
Hereditary graph property (hereditary graph class)= a class of graphs closed under deletion of vertices= a class of graphs closed under taking induced subgraphs
Formally:a set of graphs X such that
G ∈ X and H < G ⇒ H ∈ X
.
4 / 55
Hereditary Graph Properties
Hereditary graph property (Hereditary graph class)= a class of graphs closed under deletion of vertices= a class of graphs closed under taking induced subgraphs
Examples:
forests
complete graphs
line graphs
bipartite graphs
planar graphs
graphs of degree at most ∆
triangle-free graphs
perfect graphs
5 / 55
Hereditary Graph Properties
Why hereditary graph classes?
Vertex deletions are very useful for developing algorithmsfor various graph optimization problems.
Every hereditary graph property can be described in termsof forbidden induced subgraphs.
6 / 55
Hereditary Graph Properties
H-free graph = a graph that does not contain H as an inducedsubgraphFree(H) = the class of H-free graphs
Free(M) :=⋂
H∈M Free(H)M-free graph = a graph in Free(M)
Proposition
X hereditary ⇐⇒ X = Free(M) for some M
M = {all (minimal) graphs not in X}
The set M is the set of forbidden induced subgraphs for X .
7 / 55
Examples
M can be finite :
cographs= P4-free graphs
line graphs
claw-free graphs = K1,3-free graphs
triangle-free graphs = K3-free graphs
graphs of degree at most ∆
. . . or infinite :
forests = {cycles}-free graphs
bipartite graphs = {odd cycles}-free graphs
chordal graphs = {cycles of order ≥ 4}-free graphs
perfect graphs
planar graphs8 / 55
Comparing Hereditary Graph Classes
Proposition
For every two sets M1 and M2 of graphs, it holds that:
Free(M1) ⊆ Free(M2)
if and only if
(∀G2 ∈ M2)(∃G1 ∈ M1)(G1 < G2) .
ExerciseProve the above equivalence.
Example:M1 = {P4,C4},M2 = {C4,C5,C6, . . .}.
9 / 55
Recognition Problems
For a given graph class X we can define the following problem:
RECOGNITION OF GRAPHS IN XInput: A graph G.Question : Is G ∈ X?
Examples:
If X = the class of all 3-colorable graphs, the recognitionproblem is NP-complete.
If X = the class of graphs G such thatχ(G) = maxH⊆G δ(H) + 1 ,
the recognition problem is NP-complete.
If X = Free(M) where M is finite then the recognitionproblem is in P. (Why?)
10 / 55
PERFECT GRAPHS.
10 / 55
α and ω
Recall:ω(G): clique number of G = the maximum size of a clique in G.
clique = a subset of pairwise adjacent vertices
α(G): max size of an independent set in G
C is a clique in G ⇔ C is independent in G:
ω(G) = α(G)
11 / 55
Perfect Graphs
Example:ω(Kn) = n,
ω(Cn) =
{
3, if n = 3;2, otherwise.
Recall the inequality:
χ(G) ≥ ω(G) .
DefinitionA graph G is perfect, if
χ(H) = ω(H)
holds for every induced subgraph H of G.
Clearly, the class of perfect graphs is hereditary.12 / 55
Perfect Graphs
Theorem (Lov asz 1972, Perfect Graph Theorem)
A graph G is perfect if and only if its complement G is perfect.
Examples of non-perfect graphs:
odd cycles of order at least 5: C5,C7,C9, . . .
χ(C2k+1) = 3
ω(C2k+1) = 2.
their complements: C5,C7,C9, . . .
χ(C2k+1) = smallest number of pairwise disjoint cliquescovering all vertices of C2k+1 = k + 1
ω(C2k+1) = α(C2k+1) = k
13 / 55
Berge Graphs
Berge graph: a {C5,C7,C7,C9,C9, . . .}-free graph.
Claude Berge, 1926–2002, a French mathematician
He was also a sculptor,collector and expert on primitive art,founding member of the literary group Oulipo ,a Hex and chess player.
14 / 55
The Strong Perfect Graph Theorem
Berge graph: a {C5,C7,C7,C9,C9, . . .}-free graph.
Clearly, every perfect graph is Berge.
Conjecture (Berge 1963)
A graph G is perfect if and only if it is Berge.
Strong Perfect Graph Theorem (Chudnovsky, Robertson,Seymour, Thomas 2002)
A graph G is perfect if and only if it is Berge.
Total length of the proof ≈ 150 pages.
15 / 55
The Strong Perfect Graph Theorem
TheoremLet G be a Berge graph. Then either:
G belongs to a basic class; that is, either:G or G is bipartite, orG or G is the line graph of a bipartite graph, orG is a double split graph,
or G admits one of the following:a 2-join,a complement of 2-join,a balanced skew partition.
16 / 55
The Strong Perfect Graph Theorem
Why does SPGT follow from the decomposition theorem?
Suppose the SPGC is false.There is a smallest counterexample G.G is not in any of the basic classes, since those graphs areperfect.G does not admit any of the four types of decomposition sinceeach of these decompositions preserves perfectness.Contradiction.
17 / 55
Algorthmic Aspects of Perfect Graphs
Some important NP-complete graph algorithmic problems aresolvable in polynomial time for perfect graphs:
COLORABILITY,
INDEPENDENT SET,
CLIQUE.
These results are due to Grotschel-Lovasz-Schrijver (1984)and are not combinatorial.
They are based on semidefinite programming and theellipsoid method.
Existence of combinatorial algorithms is an open problem.
18 / 55
Recognizing Perfect Graphs
Theorem (Chudnovsky, Cornu ejols, Liu, Seymour, Vukovi c2005)
There is a polynomial-time algorithm for recognizing Bergegraphs.
O(|V |9)
36 pages
independent of the proof of SPGT
19 / 55
Graphs Without Odd Holes
Does the input graph contain an odd cycle?
Solvable in P.
hole: a cycle of order at least 4
Does the input graph contain an odd hole?
Open!
Theorem (Bienstock 1991)
Testing whether a graph contains an odd hole through a givenvertex is NP-complete.
20 / 55
Classes of Perfect Graphs
Some classes of perfect graphs:
bipartite graphs and their complements
line graphs of bipartite graphs (and their complements)
cographs
chordal graphs
split graphs
threshold graphs
interval graphs
21 / 55
COGRAPHS.
21 / 55
Cographs
DefinitionCographs:
K1 is a cograph
If G and H are cographs, then so is their disjoint union.
If G and H are cographs, then so is their join.
There are no further cographs.
ExerciseProve that the class of cographs is hereditary.
22 / 55
Cographs
Example:
23 / 55
Cographs
Example:
23 / 55
Cographs
Example:
23 / 55
Cographs
Example:
23 / 55
Cographs
Example:
23 / 55
Cographs
Example:
23 / 55
Cographs
Example:
23 / 55
Cographs
Example:
23 / 55
Properties of Cographs
For every cograph G 6= K1, either G or G is disconnected.
ExerciseShow that every cograph is perfect, using only the definitions ofthe two classes.
24 / 55
Properties of Cographs
The following graph is not a cograph:
Figure: P4: a self-complementary connected graph
TheoremG is a cograph if and only if G is P4-free.
Corollary
Recognition of cographs is in P.
25 / 55
Properties of Cographs
Theorem (Corneil, Perl, and Stewart 1985)
Cographs can be recognized in linear time.
The recognition algorithm uses modular decomposition.
TheoremG is a cograph if and only if G is P4-free.
26 / 55
Proof
It can be proved by induction on the number of vertices thatevery cograph is P4-free.
We prove that every P4-free graph is a cograph.
For a contradiction, let G = (V ,E) be a minimalcounterexample.(G is a P4-free graph on n vertices that is not a cograph, whileevery P4-free graph on less than n vertices is a cograph.)
Both G and G are connected.
Let x ∈ V (G). Then G − x is a cograph.
Since n > 2, we may assume that G − x is disconnected (elsereplace G with its complement).
27 / 55
Proof
Since G is connected, there exists a vertex y not adjacent to x .
Let C be the component of G − x containing y .
Since G is connected, x has a neighbor z in C.
We can then find two adjacent vertices u and v in C such thatux ∈ E and vx 6∈ E .
Let D be a component of G − x different from C.
Let w be a neighbor of x in D.
G contains an induced P4 on the vertices (v ,u, x ,w).
Contradiction.
28 / 55
Two Exercises
Exercise 1:What are the P3-free graphs?
Exercise 2:What are the bipartite P4-free graphs?
29 / 55
Cographs: Algorithmic Aspects
Proposition
The following problems are polynomially solvable for cographs:
(a) INDEPENDENT SET,
(b) CLIQUE,
(c) DOMINATING SET.
(d) COLORABILITY.
For example, α(G) can be computed recursively as follows:α(K1) = 1If K is the disjoint union of G and H then
α(K ) = α(G) + α(H) .
If K is the join of G and H then
α(K ) = max{α(G), α(H)} .30 / 55
CHORDAL GRAPHS.
30 / 55
Chordal Graphs
DefinitionA graph is chordal if every cycle on at least 4 vertices containsa chord.
chord : an edge connecting two non-consecutive vertices of thecycle.
Figure: A cycle with four chords.
31 / 55
Chordal Graphs
Example:
chordal not chordal
32 / 55
Perfectness of Chordal Graphs
A graph is chordal if and only if it is {C4,C5, . . .}-free.
Proposition
Every chordal graph is perfect.
Proof: We apply the SPGT.If a chordal graph G is not perfect thenG 6∈ Free({C5,C7,C7,C9,C9, . . .}).⇒ C2k+1 < G for some k ≥ 3.
Since C4 < C2k+1, it follows that C4 < G. Contradiction.
33 / 55
Chordal Graphs: the Intersection Model
Theorem (Gavril, 1974)
Chordal graphs are precisely the vertex-intersection graphs ofsubtrees in a tree.
34 / 55
Chordal Graphs: the Intersection Model
Theorem (Gavril, 1974)
Chordal graphs are precisely the vertex-intersection graphs ofsubtrees in a tree.
Example:
T
T1
T2 T3
T5 T4
T6
T4
T5T6
T1
T2
T3
T6
G
34 / 55
Chordal Graphs: Structural Properties
A cutset: a set of vertices X ⊆ V such that the graph G − X isdisconnected.
Theorem (Dirac, 1961)
Every minimal cutset in a chordal graph is a clique.
35 / 55
Chordal Graphs: Structural Properties
A cutset: a set of vertices X ⊆ V such that the graph G − X isdisconnected.
Theorem (Dirac, 1961)
Every minimal cutset in a chordal graph is a clique.
cutset
35 / 55
Proof
By contradiction. Suppose X is a minimal cutset in Gcontaining two non-adjacent vertices x and y .
Choose two components C and D of the (disconnected) graphG − X .
By the minimality of X , every vertex of X has a neighbor inevery component of G − X .
Let P be a shortest x-y path all of whose internal verticesbelong to C.Let Q be a shortest x-y path all of whose internal verticesbelong to D.
Then P ∪ Q is a chordless cycle on at least 4 vertices.
Contradiction.
36 / 55
Chordal Graphs: Structural Properties
A vertex is simplicial if its neighborhood is a clique.
Corollary
Let G be a chordal graph. Then,
(i) G is either complete or it contains a pair of non-adjacentsimplicial vertices.
(ii) G contains a simplicial vertex.
Theorem (Fulkerson and Gross, 1965)
A graph is chordal if and only if it has a perfect eliminationordering.
A permutation (v1, . . . , vn) of the vertices of a graph G is aperfect elimination ordering if each vi is a simplicial vertex ofG[vi , . . . , vn].
37 / 55
Chordal Graphs: Algorithmic Aspects
TheoremEvery chordal graph contains a simplicial vertex.
If G is chordal and v ∈ V (G) then G − v is chordal.
With iterative deleting of simplicial vertices, it is easy to developpolynomial time algorihtms for the following problems onchordal graphs:
CLIQUE,
COLORABILITY,
INDEPENDENT SET.
38 / 55
Chordal Graphs: Algorithmic Aspects
Suppose v is a simplicial vertex in a chordal graph G.
CLIQUE:
ω(G) = max{d(v) + 1, ω(G − v)} .
COLORABILITY:
χ(G) = max{d(v) + 1, χ(G − v)} .
Apply the greedy coloring algorithm to the vertices in thereverse of a perfect elimination ordering.
INDEPENDENT SET:
α(G) = 1 + α(G − N[v ]) .
39 / 55
SPLIT GRAPHS.
39 / 55
Definition
DefinitionA graph is split if there exists a partition of its vertex set into aclique and an independent set.
Source: http://en.wikipedia.org/wiki/Split graph
40 / 55
Forbidden Induced Subgraphs
Theorem (F oldes and Hammer, 1977)
A graph is split if and only if it is {2K2,C4,C5}-free.
2K2 C4 C5
ExerciseProve the if part of the theorem.
41 / 55
Other Properties
Corollary
A graph is split if and only if its complement is a split graph.A graph G is a split graph if and only if both G and G arechordal.
TheoremSplit graphs are precisely the vertex-intersection graphs ofsubtrees of a star.
TheoremLet d1 ≥ d2 ≥ . . . ≥ dn be the degree sequence of a graph G.Also, let m = max{i : di ≥ i − 1}. Then, G is a split graph ifand only if
∑mi=1 di = m(m − 1) +
∑ni=m+1 di .
42 / 55
Algorithmic Aspects
Split graphs can be recognized in linear time.
Other algorithmic problems on split graphs:
COLORABILITY? In P.
CLIQUE? In P.
INDEPENDENT SET? In P.
DOMINATING SET? NP-complete.
43 / 55
Open Problem . Give the forbidden induced subgraphcharacterization of graphs that can be partitioned into a cliqueand a graph of maximum degree at most 1.
44 / 55
THRESHOLD GRAPHS.
44 / 55
Definition
DefinitionA graph G = (V ,E) is threshold if there exist positive realvertex weights w(v) for all v ∈ V and a threshold t ∈ R suchthat for every vertex set X ⊆ V ,
X is independent if and only if∑
v∈X
w(v) ≤ t .
45 / 55
Definition
DefinitionA graph G = (V ,E) is threshold if there exist positive realvertex weights w(v) for all v ∈ V and a threshold t ∈ R suchthat for every vertex set X ⊆ V ,
X is independent if and only if∑
v∈X
w(v) ≤ t .
4 2
6
7
t = 7
1
45 / 55
Forbidden Induced Subgraphs
Theorem (Chv atal, Hammer 1977)
A graph is threshold if and only if it is {2K2,C4,P4}-free.
2K2 C4 P4
46 / 55
Further Characterizations
TheoremThe following properties are equivalent for a graph G:
1 G is threshold.2 G is a split cograph.3 There exist positive real vertex weights w(v) for all v ∈ V
and a threshold t ∈ R such that for every two distinctvertices u, v ∈ V,
uv ∈ E if and only if w(u) + w(v) ≥ t .
4 G can be constructed from the one-vertex graph byrepeated applications of the following two operations:
Addition of a single isolated vertex to the graph.Addition of a single dominating vertex to the graph.
47 / 55
Algorithmic Aspects
Threshold graphs can be recognized in linear time.
Other algorithmic problems on threshold graphs:
COLORABILITY? In P.
CLIQUE? In P.
INDEPENDENT SET? In P.
DOMINATING SET? In P.
48 / 55
INTERVAL GRAPHS.
48 / 55
Definition
DefinitionA graph is an interval graph if its vertices can be put intoone-to-one correspondence with a set of intervals on the realline such that two vertices are connected by an edge if and onlyif their corresponding intervals have nonempty intersection.
Source: http://en.wikipedia.org/wiki/Interval graph
49 / 55
Two Exercises
Exercise 1Prove that interval graphs are chordal.
Exercise 2Prove that the following two graphs are not interval:
50 / 55
Algorithmic Aspects
Theorem (Booth and Lueker 1976)
Interval graphs can be recognized in linear time.
Other algorithmic problems on interval graphs:
COLORABILITY? In P.
CLIQUE? In P.
INDEPENDENT SET? In P.
DOMINATING SET? In P.
51 / 55
Hasse Diagram of Some Classes of Perfect Graphs
perfect graphs
chordal bipartite
split
threshold
interval trees
cographs
52 / 55
Hasse Diagram of Some Classes of Perfect Graphs
split
tree
interval
bipartite
blockplanar bipartite
circular arc
permutation
distance-
chordal bipartite
strongly chordal
trapezoid
dually chordal
bipartite cocomparability
line graphs of
chordalAT-free
perfect
bipartite graphs
permutation
hereditary
53 / 55
What we’ll do – Week 1
1 Tue March 5: Review of basic notions in graph theory,algorithms and complexity X
2 Wed March 6: Graph colorings X
3 Thu March 7–8: Perfect graphs and their subclassesX
54 / 55
What we’ll do – Week 2
1 Tue March 19: Further examples of tractable problems,part 1
2 Wed March 20:Further examples of tractable problems, part 2Approximation algorithms for graph problems
3 Thu March 21: Lectio Magistralis lecture, “Graph classes:interrelations, structure, and algorithmic issues”
55 / 55